Machine Learning Classification Based on Individual Whole-Brain Functional Connectivity in Male OSA Patients.

Purpose: Previous studies have shown altered paired brain functional connectivity (FC) in obstructive sleep apnea (OSA) patients, linked to cognitive impairment. This study utilized individual FC analysis to investigate the distinctive FC characteristics in OSA and evaluate their classification efficiency. Methods: We included 82 moderate to severe OSA patients [41 OSA with normal cognition (OSA-NC), 41 OSA with mild cognitive impairments (OSA-MCI)] and 84 healthy control (HC). Resting-state fMRI data and clinical scale data were collected. Individual FC was derived using multi-task learning-based sparse convex alternating structure optimization, with feature selection via the least absolute shrinkage and selection operator. Support vector machine classifiers were used for OSA vs HC and OSA-NC vs OSA-MCI classification. The top 10 FC features contributing to classification were analyzed for group differences. A significance level of Results: The study results showed that individual FC achieved higher classification accuracy than traditional Pearson-based FC (OSA vs HC: 91.8% vs 79.5%; OSA-NC vs OSA-MCI: 81.3% vs 63.8%). The top 10 individual-specific FC networks contributing to classification were mainly located in the default mode network, attention network, showing significant inter-group differences in connectivity strength between the two groups. Conclusion: This study identified static individualized FC characteristics in OSA patients with varying cognitive impairments. Based on individual FC, the classification accuracy of OSA-NC and OSA-MCI was significantly improved, the individual FC may serve as a potential neuroimaging marker for predicting OSA-MCI, providing an individual clinical diagnosis and treatment evaluation.